mergekit_config.yml

merge_method: slerp  # Define the merge method at the top level

slices:
  - sources:
      - model: mattshumer/Reflection-Llama-3.1-70B
        layer_range:
          - 0
          - 40   # Adjust layer range
      - model: meta-llama/Meta-Llama-3.1-70B-Instruct
        layer_range:
          - 0
          - 40
    base_model: mattshumer/Reflection-Llama-3.1-70B  # Define the base model at the slice level

parameters:
  t:
    - filter: self_attn
      value:
        - 0.1  # Modify weights for self attention
        - 0.5
        - 0.4
        - 0.8
        - 1
    - filter: mlp
      value:
        - 0.9  # Modify weights for MLP layers
        - 0.6
        - 0.7
        - 0.4
        - 0.2
    - value: 0.7  # General merge weight

dtype: bfloat16  # Keep for TPU efficiency